Jiang Episode Publisher

Jiang Episode Publisher

Use this after content/lens/episodes/<source-slug>/read.json exists. This skill is mechanical publication and validation. It should not become the episode writer.

Inputs

• content/sources/videos/<source-slug>/
• content/lens/evidence/videos/<source-slug>.semantic.json
• content/lens/episodes/<source-slug>/read.json
• compiler scripts under ops/scripts/

Flow

1. Compile content:

node ops/scripts/compile-content.mjs

2. Validate content:

node ops/scripts/validate-content.mjs

3. Build the website:

cd website && npm run build

4. Verify expected outputs:

website/src/data/lens/episodes/<source-slug>.json
/episodes/<source-slug>/
/episodes/<source-slug>/transcript/

# or, when source_class is interview:
website/src/data/lens/interviews/<source-slug>.json
/interviews/<source-slug>/
/interviews/<source-slug>/transcript/

5. If a dev server is running, spot-check the episode and transcript routes. For UI changes, take screenshots before accepting.

Checks

Confirm:

• the episode or interview page renders,
• the transcript page renders,
• the video ID is present,
• timestamp links seek the video,
• transcript current-segment highlighting works,
• phrase-level highlighting exists only when timed_chunks exist,
• evidence marks and lens point hovers render without broken refs,
• generated data is fresh.

Boundary

Do not rewrite public prose except for small structural fixes required by validation. If the read is shallow, wrong, or missing Jiang's voice, hand it back to jiang-episode-read-writer.

Do not edit generated files by hand.

After successful publication, stop. Corpus impact, lens links, concept extension, new lens seeds, atlas mutation, and canon-candidate review are downstream jobs for separate agents. Mention obvious follow-up in the handoff, but do not create or require corpus-impact.json as part of episode publication.